Advantage of clustering algorithm

Question

We don't need the name of attributes in clustering, so, if I do not know my attribute names how can I understand that which data should I enter and also if I do not know the name of the attributes how can I give the axis name of the plotted graph? If it is the advantage of clustering algorithm, then why it is not the advantage of other algorithms, because if we do not know the name of the attributes we can create our models because if we have attribute values that is enough, but why we need attribute names too in other algorithms?

Answer 1

@Sanniddha Chakrabarti Thanks, If the products data is not already labeled and ready for training, you can start with a clustering problem.
Once you identify those clusters, a Domain Expert can review those clusters and try to set a name for each (the Categories/classes).
Then, all the data can be labeled according to those new categories/classes and finally train a model.

From there, with a trained model, you could “predict” what category/class a product should be assigned based on its product’s name and description.

• For the clustering problem you need to directly use a framework such as Scikit-Learn (or even ML.NET in C#).
• Defining a category/class name for each identified cluster needs to be done manually by a Domain Expert.
• Labeling the data could be semi-automated with a custom program, based on the multiple clusters defined, it would label each row with a new class-caterogy-name defined for each of the identified clusters.
• For the multi-class classification model training, you can use Azure Automated ML as the easiest approach.